Abstract: Based on differential-geometry features of edge curves, a new data segmentation method is proposed to segment the unorganized noise point-cloud. An algorithm revised from the TAUBIN’s paper is put forward first to estimate the principle curvatures and principle directions of the unorganized noise points. By analyze the variability of curvature in principle di-rection for each point, the G1 or G2 continuous edge-points can be detected. The acquired edge-points form into edge stripes, which segment the point-cloud into a few sub-regions. Finally a region-growing way is adopted to identify every sub-area. Results indicate that the presented method can overcome noise influence and recognize the G1 and G2 edges of unorganized noise point-cloud effectively, and the method can directly ac-quire good G2 edges of the complicated object.

Key words: Curvature estimation, Data segmentation, Unorganized noise point-cloud